tests/test_urcu_lfq.c

   1 /*
   2  * test_urcu_lfq.c
   3  *
   4  * Userspace RCU library - example RCU-based lock-free queue
   5  *
   6  * Copyright February 2010 - Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
   7  * Copyright February 2010 - Paolo Bonzini <pbonzini@redhat.com>
   8  *
   9  * This program is free software; you can redistribute it and/or modify
  10  * it under the terms of the GNU General Public License as published by
  11  * the Free Software Foundation; either version 2 of the License, or
  12  * (at your option) any later version.
  13  *
  14  * This program is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  17  * GNU General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU General Public License along
  20  * with this program; if not, write to the Free Software Foundation, Inc.,
  21  * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
  22  */
  23
  24 #define _GNU_SOURCE
  25 #include "../config.h"
  26 #include <stdio.h>
  27 #include <pthread.h>
  28 #include <stdlib.h>
  29 #include <stdint.h>
  30 #include <stdbool.h>
  31 #include <string.h>
  32 #include <sys/types.h>
  33 #include <sys/wait.h>
  34 #include <unistd.h>
  35 #include <stdio.h>
  36 #include <assert.h>
  37 #include <errno.h>
  38
  39 #include <urcu/arch.h>
  40 #include <urcu/tls-compat.h>
  41 #include "cpuset.h"
  42
  43 #ifdef __linux__
  44 #include <syscall.h>
  45 #endif
  46
  47 /* hardcoded number of CPUs */
  48 #define NR_CPUS 16384
  49
  50 #if defined(_syscall0)
  51 _syscall0(pid_t, gettid)
  52 #elif defined(__NR_gettid)
  53 static inline pid_t gettid(void)
  54 {
  55         return syscall(__NR_gettid);
  56 }
  57 #else
  58 #warning "use pid as tid"
  59 static inline pid_t gettid(void)
  60 {
  61         return getpid();
  62 }
  63 #endif
  64
  65 #ifndef DYNAMIC_LINK_TEST
  66 #define _LGPL_SOURCE
  67 #endif
  68 #include <urcu.h>
  69 #include <urcu/cds.h>
  70
  71 static volatile int test_go, test_stop;
  72
  73 static unsigned long rduration;
  74
  75 static unsigned long duration;
  76
  77 /* read-side C.S. duration, in loops */
  78 static unsigned long wdelay;
  79
  80 static inline void loop_sleep(unsigned long loops)
  81 {
  82         while (loops-- != 0)
  83                 caa_cpu_relax();
  84 }
  85
  86 static int verbose_mode;
  87
  88 #define printf_verbose(fmt, args...)            \
  89         do {                                    \
  90                 if (verbose_mode)               \
  91                         printf(fmt, args);      \
  92         } while (0)
  93
  94 static unsigned int cpu_affinities[NR_CPUS];
  95 static unsigned int next_aff = 0;
  96 static int use_affinity = 0;
  97
  98 pthread_mutex_t affinity_mutex = PTHREAD_MUTEX_INITIALIZER;
  99
 100 static void set_affinity(void)
 101 {
 102 #if HAVE_SCHED_SETAFFINITY
 103         cpu_set_t mask;
 104         int cpu, ret;
 105 #endif /* HAVE_SCHED_SETAFFINITY */
 106
 107         if (!use_affinity)
 108                 return;
 109
 110 #if HAVE_SCHED_SETAFFINITY
 111         ret = pthread_mutex_lock(&affinity_mutex);
 112         if (ret) {
 113                 perror("Error in pthread mutex lock");
 114                 exit(-1);
 115         }
 116         cpu = cpu_affinities[next_aff++];
 117         ret = pthread_mutex_unlock(&affinity_mutex);
 118         if (ret) {
 119                 perror("Error in pthread mutex unlock");
 120                 exit(-1);
 121         }
 122
 123         CPU_ZERO(&mask);
 124         CPU_SET(cpu, &mask);
 125 #if SCHED_SETAFFINITY_ARGS == 2
 126         sched_setaffinity(0, &mask);
 127 #else
 128         sched_setaffinity(0, sizeof(mask), &mask);
 129 #endif
 130 #endif /* HAVE_SCHED_SETAFFINITY */
 131 }
 132
 133 /*
 134  * returns 0 if test should end.
 135  */
 136 static int test_duration_dequeue(void)
 137 {
 138         return !test_stop;
 139 }
 140
 141 static int test_duration_enqueue(void)
 142 {
 143         return !test_stop;
 144 }
 145
 146 static DEFINE_URCU_TLS(unsigned long long, nr_dequeues);
 147 static DEFINE_URCU_TLS(unsigned long long, nr_enqueues);
 148
 149 static DEFINE_URCU_TLS(unsigned long long, nr_successful_dequeues);
 150 static DEFINE_URCU_TLS(unsigned long long, nr_successful_enqueues);
 151
 152 static unsigned int nr_enqueuers;
 153 static unsigned int nr_dequeuers;
 154
 155 struct test {
 156         struct cds_lfq_node_rcu list;
 157         struct rcu_head rcu;
 158 };
 159
 160 static struct cds_lfq_queue_rcu q;
 161
 162 void *thr_enqueuer(void *_count)
 163 {
 164         unsigned long long *count = _count;
 165
 166         printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n",
 167                         "enqueuer", (unsigned long) pthread_self(),
 168                         (unsigned long) gettid());
 169
 170         set_affinity();
 171
 172         rcu_register_thread();
 173
 174         while (!test_go)
 175         {
 176         }
 177         cmm_smp_mb();
 178
 179         for (;;) {
 180                 struct test *node = malloc(sizeof(*node));
 181                 if (!node)
 182                         goto fail;
 183                 cds_lfq_node_init_rcu(&node->list);
 184                 rcu_read_lock();
 185                 cds_lfq_enqueue_rcu(&q, &node->list);
 186                 rcu_read_unlock();
 187                 URCU_TLS(nr_successful_enqueues)++;
 188
 189                 if (caa_unlikely(wdelay))
 190                         loop_sleep(wdelay);
 191 fail:
 192                 URCU_TLS(nr_enqueues)++;
 193                 if (caa_unlikely(!test_duration_enqueue()))
 194                         break;
 195         }
 196
 197         rcu_unregister_thread();
 198
 199         count[0] = URCU_TLS(nr_enqueues);
 200         count[1] = URCU_TLS(nr_successful_enqueues);
 201         printf_verbose("enqueuer thread_end, thread id : %lx, tid %lu, "
 202                        "enqueues %llu successful_enqueues %llu\n",
 203                        pthread_self(),
 204                         (unsigned long) gettid(),
 205                        URCU_TLS(nr_enqueues), URCU_TLS(nr_successful_enqueues));
 206         return ((void*)1);
 207
 208 }
 209
 210 static
 211 void free_node_cb(struct rcu_head *head)
 212 {
 213         struct test *node =
 214                 caa_container_of(head, struct test, rcu);
 215         free(node);
 216 }
 217
 218 void *thr_dequeuer(void *_count)
 219 {
 220         unsigned long long *count = _count;
 221
 222         printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n",
 223                         "dequeuer", (unsigned long) pthread_self(),
 224                         (unsigned long) gettid());
 225
 226         set_affinity();
 227
 228         rcu_register_thread();
 229
 230         while (!test_go)
 231         {
 232         }
 233         cmm_smp_mb();
 234
 235         for (;;) {
 236                 struct cds_lfq_node_rcu *qnode;
 237
 238                 rcu_read_lock();
 239                 qnode = cds_lfq_dequeue_rcu(&q);
 240                 rcu_read_unlock();
 241
 242                 if (qnode) {
 243                         struct test *node;
 244
 245                         node = caa_container_of(qnode, struct test, list);
 246                         call_rcu(&node->rcu, free_node_cb);
 247                         URCU_TLS(nr_successful_dequeues)++;
 248                 }
 249
 250                 URCU_TLS(nr_dequeues)++;
 251                 if (caa_unlikely(!test_duration_dequeue()))
 252                         break;
 253                 if (caa_unlikely(rduration))
 254                         loop_sleep(rduration);
 255         }
 256
 257         rcu_unregister_thread();
 258         printf_verbose("dequeuer thread_end, thread id : %lx, tid %lu, "
 259                        "dequeues %llu, successful_dequeues %llu\n",
 260                        pthread_self(),
 261                         (unsigned long) gettid(),
 262                        URCU_TLS(nr_dequeues), URCU_TLS(nr_successful_dequeues));
 263         count[0] = URCU_TLS(nr_dequeues);
 264         count[1] = URCU_TLS(nr_successful_dequeues);
 265         return ((void*)2);
 266 }
 267
 268 void test_end(struct cds_lfq_queue_rcu *q, unsigned long long *nr_dequeues)
 269 {
 270         struct cds_lfq_node_rcu *snode;
 271
 272         do {
 273                 snode = cds_lfq_dequeue_rcu(q);
 274                 if (snode) {
 275                         struct test *node;
 276
 277                         node = caa_container_of(snode, struct test, list);
 278                         free(node);     /* no more concurrent access */
 279                         (*nr_dequeues)++;
 280                 }
 281         } while (snode);
 282 }
 283
 284 void show_usage(int argc, char **argv)
 285 {
 286         printf("Usage : %s nr_dequeuers nr_enqueuers duration (s) <OPTIONS>\n",
 287                 argv[0]);
 288         printf("OPTIONS:\n");
 289         printf("        [-d delay] (enqueuer period (in loops))\n");
 290         printf("        [-c duration] (dequeuer period (in loops))\n");
 291         printf("        [-v] (verbose output)\n");
 292         printf("        [-a cpu#] [-a cpu#]... (affinity)\n");
 293         printf("\n");
 294 }
 295
 296 int main(int argc, char **argv)
 297 {
 298         int err;
 299         pthread_t *tid_enqueuer, *tid_dequeuer;
 300         void *tret;
 301         unsigned long long *count_enqueuer, *count_dequeuer;
 302         unsigned long long tot_enqueues = 0, tot_dequeues = 0;
 303         unsigned long long tot_successful_enqueues = 0,
 304                            tot_successful_dequeues = 0;
 305         unsigned long long end_dequeues = 0;
 306         int i, a;
 307
 308         if (argc < 4) {
 309                 show_usage(argc, argv);
 310                 return -1;
 311         }
 312
 313         err = sscanf(argv[1], "%u", &nr_dequeuers);
 314         if (err != 1) {
 315                 show_usage(argc, argv);
 316                 return -1;
 317         }
 318
 319         err = sscanf(argv[2], "%u", &nr_enqueuers);
 320         if (err != 1) {
 321                 show_usage(argc, argv);
 322                 return -1;
 323         }
 324
 325         err = sscanf(argv[3], "%lu", &duration);
 326         if (err != 1) {
 327                 show_usage(argc, argv);
 328                 return -1;
 329         }
 330
 331         for (i = 4; i < argc; i++) {
 332                 if (argv[i][0] != '-')
 333                         continue;
 334                 switch (argv[i][1]) {
 335                 case 'a':
 336                         if (argc < i + 2) {
 337                                 show_usage(argc, argv);
 338                                 return -1;
 339                         }
 340                         a = atoi(argv[++i]);
 341                         cpu_affinities[next_aff++] = a;
 342                         use_affinity = 1;
 343                         printf_verbose("Adding CPU %d affinity\n", a);
 344                         break;
 345                 case 'c':
 346                         if (argc < i + 2) {
 347                                 show_usage(argc, argv);
 348                                 return -1;
 349                         }
 350                         rduration = atol(argv[++i]);
 351                         break;
 352                 case 'd':
 353                         if (argc < i + 2) {
 354                                 show_usage(argc, argv);
 355                                 return -1;
 356                         }
 357                         wdelay = atol(argv[++i]);
 358                         break;
 359                 case 'v':
 360                         verbose_mode = 1;
 361                         break;
 362                 }
 363         }
 364
 365         printf_verbose("running test for %lu seconds, %u enqueuers, "
 366                        "%u dequeuers.\n",
 367                        duration, nr_enqueuers, nr_dequeuers);
 368         printf_verbose("Writer delay : %lu loops.\n", rduration);
 369         printf_verbose("Reader duration : %lu loops.\n", wdelay);
 370         printf_verbose("thread %-6s, thread id : %lx, tid %lu\n",
 371                         "main", (unsigned long) pthread_self(),
 372                         (unsigned long) gettid());
 373
 374         tid_enqueuer = malloc(sizeof(*tid_enqueuer) * nr_enqueuers);
 375         tid_dequeuer = malloc(sizeof(*tid_dequeuer) * nr_dequeuers);
 376         count_enqueuer = malloc(2 * sizeof(*count_enqueuer) * nr_enqueuers);
 377         count_dequeuer = malloc(2 * sizeof(*count_dequeuer) * nr_dequeuers);
 378         cds_lfq_init_rcu(&q, call_rcu);
 379         err = create_all_cpu_call_rcu_data(0);
 380         if (err) {
 381                 printf("Per-CPU call_rcu() worker threads unavailable. Using default global worker thread.\n");
 382         }
 383
 384         next_aff = 0;
 385
 386         for (i = 0; i < nr_enqueuers; i++) {
 387                 err = pthread_create(&tid_enqueuer[i], NULL, thr_enqueuer,
 388                                      &count_enqueuer[2 * i]);
 389                 if (err != 0)
 390                         exit(1);
 391         }
 392         for (i = 0; i < nr_dequeuers; i++) {
 393                 err = pthread_create(&tid_dequeuer[i], NULL, thr_dequeuer,
 394                                      &count_dequeuer[2 * i]);
 395                 if (err != 0)
 396                         exit(1);
 397         }
 398
 399         cmm_smp_mb();
 400
 401         test_go = 1;
 402
 403         for (i = 0; i < duration; i++) {
 404                 sleep(1);
 405                 if (verbose_mode)
 406                         write (1, ".", 1);
 407         }
 408
 409         test_stop = 1;
 410
 411         for (i = 0; i < nr_enqueuers; i++) {
 412                 err = pthread_join(tid_enqueuer[i], &tret);
 413                 if (err != 0)
 414                         exit(1);
 415                 tot_enqueues += count_enqueuer[2 * i];
 416                 tot_successful_enqueues += count_enqueuer[2 * i + 1];
 417         }
 418         for (i = 0; i < nr_dequeuers; i++) {
 419                 err = pthread_join(tid_dequeuer[i], &tret);
 420                 if (err != 0)
 421                         exit(1);
 422                 tot_dequeues += count_dequeuer[2 * i];
 423                 tot_successful_dequeues += count_dequeuer[2 * i + 1];
 424         }
 425
 426         test_end(&q, &end_dequeues);
 427         err = cds_lfq_destroy_rcu(&q);
 428         assert(!err);
 429
 430         printf_verbose("total number of enqueues : %llu, dequeues %llu\n",
 431                        tot_enqueues, tot_dequeues);
 432         printf_verbose("total number of successful enqueues : %llu, "
 433                        "successful dequeues %llu\n",
 434                        tot_successful_enqueues, tot_successful_dequeues);
 435         printf("SUMMARY %-25s testdur %4lu nr_enqueuers %3u wdelay %6lu "
 436                 "nr_dequeuers %3u "
 437                 "rdur %6lu nr_enqueues %12llu nr_dequeues %12llu "
 438                 "successful enqueues %12llu successful dequeues %12llu "
 439                 "end_dequeues %llu nr_ops %12llu\n",
 440                 argv[0], duration, nr_enqueuers, wdelay,
 441                 nr_dequeuers, rduration, tot_enqueues, tot_dequeues,
 442                 tot_successful_enqueues,
 443                 tot_successful_dequeues, end_dequeues,
 444                 tot_enqueues + tot_dequeues);
 445         if (tot_successful_enqueues != tot_successful_dequeues + end_dequeues)
 446                 printf("WARNING! Discrepancy between nr succ. enqueues %llu vs "
 447                        "succ. dequeues + end dequeues %llu.\n",
 448                        tot_successful_enqueues,
 449                        tot_successful_dequeues + end_dequeues);
 450
 451         free_all_cpu_call_rcu_data();
 452         free(count_enqueuer);
 453         free(count_dequeuer);
 454         free(tid_enqueuer);
 455         free(tid_dequeuer);
 456         return 0;
 457 }